Generative AI · Best Practices Guide

Talk to AI
like a pro.

Most people use AI like a search engine — vague, hopeful, disappointed. This guide teaches you six principles that transform mediocre prompts into powerful instructions that get exactly what you need.

Give Direction Specify Format Provide Examples Evaluate Quality Divide Labor Prompt Formula
Scroll to learn
01 — Give Direction

Tell AI who to be
and what to know.

AI responds better when it has a role to play, context to work from, and clear rules to follow. Don't just ask — direct.

🎭
Assign a Role
Start your prompt by telling AI who it is. A financial analyst responds differently than a creative writer — even to the same question.
"Act as a senior investment banker reviewing this pitch deck." "You are a plain-language translator for non-technical readers." "Take the role of a strict editor focused on clarity and brevity."
🔍
Warm It Up First
Before asking AI to do a task, ask it to recall what it knows about the topic. This "prewarming" surfaces relevant context before the real request.
"What do you know about DCF analysis?" → "Using that, build me a model outline." "Summarize best practices for cold emails." → "Now write one for my use case."
📋
Set the Context
Tell AI who the output is for and what the situation is. Context shapes tone, depth, and assumptions automatically.
"This report is intended for a non-finance executive audience." "Read the attached data and respond as a financial advisor."
🚦
Write the Rules
Eliminate guesswork with explicit do's and don'ts. AI will follow constraints when you state them clearly — it won't infer what you didn't say.
"Do not use jargon. Keep every sentence under 20 words." "Always cite your sources. Do not fabricate statistics." "Use an encouraging tone. Avoid negative framing."
02 — Specify Format

Define the shape
of your answer.

If you don't describe what the output should look like, AI will guess — and it often guesses wrong. Tell it exactly what you want.

📐
Output Type
Name the format: bullets, numbered list, table, JSON, outline, code, quiz, paragraph, email, slide deck. The more specific, the better.
"Return as a table with columns: Risk | Likelihood | Mitigation." "Output as a 5-bullet executive summary." "Create a one-file static webpage with HTML, CSS, and JS."
📏
Length & Scope
Specify how long or short the output should be. AI defaults to verbose — unless you constrain it.
"Keep each bullet to ≤ 12 words." "Write a 150-word summary — no more." "Give me 3 options, not a full explanation."
🎯
Tone & Audience
Who is reading this? Match the tone to the reader. The same content for a CEO and a college student should sound completely different.
"Write for a first-generation college student with no finance background." "Use executive-level language. Assume the reader has 20 years of experience."
🔒
Constraints
Add hard limits that keep output useful: no jargon, include steps, cite formulas, no filler phrases. Think like an editor before you prompt.
"Do not include an intro paragraph. Start with the answer." "Include at least one real-world example per point."
03 — Provide Examples

Show, don't
just tell.

Examples reduce what the AI has to assume. A prompt with no examples is a guessing game. See the difference a few words make.

❌ Weak Prompt
Explain DCF.
Too vague — no audience, no depth, no format. You'll get a textbook definition that may be useless for your actual need.
Upgraded Prompt
✓ Strong Prompt
You are a finance professor teaching undergrads. Explain DCF analysis in plain English — no formulas. Use a real-world example involving a coffee shop. Keep it under 200 words and end with one sentence on when NOT to use DCF.
Role assigned ✓ · Audience defined ✓ · Format specified ✓ · Example requested ✓ · Length constrained ✓ · Constraint added ✓
❌ Weak Prompt
Write me a follow-up email.
Who is it to? After what? What's the goal? AI has no choice but to invent a generic email that probably misses the mark entirely.
Upgraded Prompt
✓ Strong Prompt
Write a follow-up email to a Morgan Stanley recruiter after a first-round interview for a Wealth Management Analyst role. I interviewed 3 days ago. Tone: professional but warm, not stiff. Length: 5–7 sentences. Do not use the phrase "I hope this email finds you well." End with a specific question about next steps.
Context provided ✓ · Tone specified ✓ · Length constrained ✓ · Rule added (no cliché) ✓ · Clear ending goal ✓
❌ Weak Prompt
Tell me about the stock market.
Impossibly broad. The AI could write 10,000 words and still not answer what you actually need. No scope, no angle, no purpose.
Upgraded Prompt
✓ Strong Prompt
You are a financial analyst briefing a new intern. In 3 bullet points, explain how rising interest rates affect equity valuations. Use simple language, one analogy, and no more than 2 sentences per bullet. Assume the reader has zero economics background.
Role assigned ✓ · Scope narrowed ✓ · Format explicit ✓ · Analogy requested ✓ · Audience defined ✓
❌ Weak Prompt
Fix my resume bullet.
Fix it how? For what role? What's wrong with it? AI will make minor edits with no strategic direction — you'll get polish, not impact.
Upgraded Prompt
✓ Strong Prompt
Rewrite this resume bullet for a Wealth Management Analyst role at a bulge bracket bank. Make it results-oriented using the formula: [Action verb] + [what I did] + [measurable outcome]. Do not add metrics I haven't provided. Keep it under 20 words. Original: "Helped with a project about lending."
Goal clear ✓ · Formula provided ✓ · Constraint set ✓ · Honesty rule added ✓ · Original given ✓
The Prompt Formula

Every great prompt
has four parts.

Think of this as your template. You don't always need all four — but the more you include, the less AI has to assume.

Role + Context + Format + Example = Great Output
Role
Who should AI be? Assign expertise, perspective, or persona.
"You are a senior equity analyst..."
Context
What's the situation? Who is the audience? What's the goal?
"...reviewing this pitch for a Series A investor..."
Format
What should the output look like? Length, structure, tone, constraints.
"...in 5 bullets, plain English, under 150 words."
Example
Show one example of what good looks like — or what to avoid.
"Here's a strong bullet: [X]. Write in this style."
04 — Evaluate Quality

Don't deploy blindly.
Test and iterate.

Sending a prompt once and accepting the first result is like publishing a first draft. Quality prompting is a feedback loop.

🚫
Avoid Blind Prompting
Deploying prompts without checking their output leads to unpredictable — and sometimes embarrassing — results. Always read what AI produces before using it.
🧪
Task Shapes the Model
Different AI models are better at different things. Test your prompt across models if the stakes are high. Don't assume one model is best for every job.
⚙️
Build Quality In
Add quality guidelines directly to your prompt: "Flag any assumptions you've made," "If unsure, say so," "Prioritize accuracy over completeness."
👁️
Keep Humans in the Loop
Automated outputs can miss nuance, cultural context, and factual errors that only a human catches. Review what AI gives you — especially before sharing.
05 — Divide Labor

One AI, one job.
Not everything at once.

Loading a single AI with too many tasks in one conversation bloats the context, raises costs, and makes responses less reliable. Break big tasks apart.

❌ One overloaded prompt
The "Do Everything" Mistake
"Research my competitors, write a SWOT analysis, draft a pitch deck outline, create 10 cold email templates, and suggest a pricing strategy."

Too many tasks = shallow answers on all of them.
→
Prompt 1
Research my top 3 competitors and summarize their pricing models.
Prompt 2
Using that research, write a SWOT analysis focused on pricing gaps.
Prompt 3
Based on the SWOT, draft a pitch deck outline for investors.
💡 The rule: Each prompt should have one clear goal. When a task grows complex, split it into a chain — where each prompt builds on the output of the last. This is called prompt chaining, and it's how professionals get consistently great results.
06 — Quick Reference Checklist

Before you hit send,
ask yourself:

Run through this before every important prompt. It takes 30 seconds and saves hours of back-and-forth.

01
Did I assign a role?
AI performs better with a persona. Even "Act as an expert in X" makes a difference.
02
Did I give context?
Who is the audience? What's the goal? What does AI need to know to help properly?
03
Did I specify format?
Bullets, table, email, code, essay? What length? What tone? Define the shape of your answer.
04
Did I add an example?
Even one example of what "good" looks like dramatically improves the output quality.
05
Did I set the rules?
What should AI never do? What must it always include? Don't assume — state it explicitly.
06
Is this one job?
If the prompt has more than one major task, split it. One goal per prompt = sharper results.